Automorphic Equivalence-aware Graph Neural Network

This repository holds an implementation of Automorphic Equivalence-aware Graph Neural Network NeurIPS 2021.

Dependencies

This implementation depends on the following enviornment and packages:

python == 2.7
pytorch == 1.4
numpy == 1.17.2
scipy == 1.6.3
sklearn ==0.24.1
h5py == 2.9.0
GPUtil ==1.4.0
setproctitle == 1.1.10

Training and Evaluation

To train and evaluate the model(s) in the paper, run this command:

python grape_model.py --data='cite|cora' --gpu='0' --lr=0.003 --wd=0.00003 --dropout=0.5 --hid=32 

Results

Our model achieves the following classification accuracy:

	Hamilton	Lehigh	Rochester	JHU	Amherst	Cora	Citeseer	Amazon
GRAPE	28.1%	27.3%	25.0%	34.6%	32.6%	87.1%	74.6%	58.6%